Following the rule of human-centricity, human motion perception is the basis for constructing human motion digital twin. In particular, accurate motion pose calculation and pose calibration are the keys to reconstruct human motion. When capturing human data based on inertial motion capture system, it is still challenging to model motion and minimize noise. At the same time, the development of convenient and effective posture calibration methods is a prerequisite for promoting human motion digital twin. This work proposes an inertial motion capture system for human motion digital twin. On the basis of the developed system, this work focuses on the attitude calculation method of adaptive extended Kalman filter and the attitude calibration method based on kinematic constraints. The proposed system achieves an average root mean square error of 4.7° in estimating orientations in comparison with an optical motion capture system. Experimental results show a great correlation (92.5%) between the proposed system and the optical system, and satisfy the requirement that over 95% of data points reside within the consistency interval. The abilities of the proposed system focus on human-centric applications based on the integration of human, cyber system, and physical system, such as motion monitoring, human–robot teleoperation.